基于有限时间对偶神经网络的冗余机械臂重复运动规划

Finite-time dual neural network for solving repetitive motion planning of redundant manipulator

在考虑关节物理极限的情况下,将冗余机械臂的逆运动学解析问题抽象为带约束的重复运动规划(RMP)方案.针对速度层的带约束RMP方案,本文提出了一种新型的递归神经网络,即有限时间对偶神经网络(FTDNN),用以求解该类带约束RMP方案.相比于传统的递归神经网络,该FTDNN模型具有有限时间收敛特性,不仅能够改进收敛的速度,并且能够获得较高的收敛精度.通过李雅普诺夫稳定性定理验证了FTDNN模型的渐近稳定性,并进一步计算出FTDNN模型求解带约束RMP方案最优解的时间上界.基于冗余机械臂PA10的计算仿真结果验证了FTDNN模型求解带约束RMP方案的有效性和可行性.最后在Dobot Magician实物机械臂上的实验结果表明本文提出的有限时间对偶神经网络方法可以有效实现机械臂各关节角的重复运动.

Resolution of the inverse-kinematics redundancy problem for redundant manipulators can be transformed into a repetitive motion planning(RMP)problem by incorporating joint-angle limits and joint-velocity limits.In this paper,a new recurrent neural network,finite-time dual neural network(FTDNN),is proposed for solving this type of constrained RMP problem in velocity layer.Compared with the traditional recurrent neural networks,the FTDNN model has finite-time convergence characteristics,which not only improves the convergent speed but also achieves higher convergent accuracy.Furthermore,asymptotic stability of the FTDNN model is verified by the Lyapunov’s stability theorem,and the time upper bound for the optimal solution of the FTDNN model with constrained RMP problem is presented.Computational simulation results based on the redundant manipulator PA10 demonstrate the validity and feasibility of the FTDNN model for solving the constrained RMP problem.Finally,experimental results on the Dobot Magician manipulator show that the proposed FTDNN model can effectively realize the repetitive kinematics of each joint angle.